@piagent/platform 1.6.0 → 1.7.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +104 -0
- package/README.md +24 -15
- package/README.vi.md +36 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +27 -3
- package/benchmarks/production-v1/suite.json +5 -1
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +16 -16
- package/docs/authority-manifest.md +1 -1
- package/docs/auto-delegation-policy.md +38 -73
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +30 -35
- package/docs/context-engine.md +5 -3
- package/docs/context-window-policy.md +37 -1
- package/docs/distribution-standard.md +9 -9
- package/docs/en/architecture.md +26 -13
- package/docs/languages.json +4 -4
- package/docs/operator-manual-vietnamese.md +26 -46
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +7 -0
- package/docs/quality-benchmark.md +299 -23
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +32 -9
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/runtime-harness-standard.md +3 -3
- package/docs/security-threat-model.md +1 -1
- package/docs/subagent-orchestration-capabilities.md +19 -51
- package/docs/subagents-and-multiagent.md +21 -47
- package/docs/task-implementation-contract.md +3 -3
- package/docs/task-lifecycle-vietnamese.md +1 -1
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs/vi/architecture.md +26 -13
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +89 -95
- package/docs-site/commands.html +1 -1
- package/docs-site/content/benchmark.html +88 -94
- package/docs-site/content/en/benchmark.html +134 -90
- package/docs-site/content/en/index.html +19 -19
- package/docs-site/content/en/runtime.html +5 -5
- package/docs-site/content/en/whats-new.html +23 -1
- package/docs-site/content/en/workflows.html +2 -2
- package/docs-site/content/index.html +19 -19
- package/docs-site/content/runtime.html +6 -6
- package/docs-site/content/whats-new.html +23 -1
- package/docs-site/content/workflows.html +2 -2
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +135 -91
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +20 -20
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +6 -6
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -2
- package/docs-site/en/workflows.html +3 -3
- package/docs-site/index.html +20 -20
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +7 -7
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -2
- package/docs-site/workflows.html +3 -3
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
- package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
- package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/helper-request.invalid.json +1 -1
- package/evals/fixtures/helper-request.valid.json +1 -1
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
- package/evals/fs4-readiness-matrix.v1.json +2 -2
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +10 -6
- package/evals/fs5-pilot-protocol.v1.json +3 -3
- package/evals/fs5-pilot-protocol.v2.json +8 -8
- package/evals/fs5-pilot-protocol.v3.json +8 -8
- package/evals/fs5-pilot-protocol.v4.json +7 -7
- package/evals/fs5-pilot-protocol.v5.json +10 -8
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/evals/runtime-conformance-v1/README.md +43 -0
- package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
- package/evals/runtime-conformance-v1/lane.json +57 -0
- package/evals/runtime-conformance-v1/paid-canary.json +31 -0
- package/evals/runtime-conformance-v1/runner.mjs +456 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +15 -6
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +12 -17
- package/packages/piagent-core/extensions/context-index-policy.js +4 -0
- package/packages/piagent-core/extensions/context-query-paths.js +57 -0
- package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
- package/packages/piagent-core/extensions/criterion-graph.js +4 -1
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
- package/packages/piagent-core/extensions/guard-types.ts +9 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +140 -43
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +43 -2
- package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
- package/packages/piagent-core/prompts/be-to-fe.md +2 -8
- package/packages/piagent-core/prompts/plan.md +1 -1
- package/packages/piagent-core/prompts/platform-improve.md +2 -6
- package/packages/piagent-core/prompts/review.md +2 -7
- package/packages/piagent-core/prompts/scout.md +1 -1
- package/packages/piagent-core/prompts/task.md +4 -3
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
- package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
- package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
- package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
- package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
- package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
- package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
- package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
- package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
- package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
- package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
- package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
- package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
- package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
- package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
- package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
- package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
- package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
- package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
- package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
- package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
- package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
- package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
- package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
- package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
- package/packages/piagent-core/subagents/piagent-planner.md +6 -4
- package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
- package/packages/piagent-core/subagents/piagent-scout.md +4 -2
- package/packages/piagent-core/subagents/piagent-worker.md +6 -4
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
- package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
- package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
- package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
- package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
- package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +150 -11
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/project-profile.schema.json +1 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +45 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +224 -387
- package/scripts/benchmark-runner-finalization.mjs +144 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-subagents.sh +42 -34
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/helper-evaluation.mjs +1 -1
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/team-doctor.sh +1 -1
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +41 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
|
@@ -0,0 +1,242 @@
|
|
|
1
|
+
import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
|
|
2
|
+
|
|
3
|
+
export const PIAGENT_FAST_MODE_STATE_ENTRY_TYPE = "piagent-fast-mode-state";
|
|
4
|
+
export const PIAGENT_SERVICE_TIER_RECEIPT_ENTRY_TYPE = "piagent-service-tier-receipt";
|
|
5
|
+
export const PIAGENT_FAST_SERVICE_TIER = "fast" as const;
|
|
6
|
+
export const PIAGENT_FAST_PROVIDER_SERVICE_TIER = "priority" as const;
|
|
7
|
+
|
|
8
|
+
type FastModeSource = "default" | "environment" | "session-command" | "session-state";
|
|
9
|
+
|
|
10
|
+
export type ServiceTierReceipt = {
|
|
11
|
+
schemaVersion: 1;
|
|
12
|
+
enabled: boolean;
|
|
13
|
+
mode: "standard" | "fast";
|
|
14
|
+
source: FastModeSource;
|
|
15
|
+
requestedServiceTier: "fast" | null;
|
|
16
|
+
observedRequestServiceTier: "priority" | null;
|
|
17
|
+
providerResponseServiceTier: null;
|
|
18
|
+
providerResponseEvidence: "unavailable-host-api";
|
|
19
|
+
provider: string | null;
|
|
20
|
+
modelId: string | null;
|
|
21
|
+
applied: boolean;
|
|
22
|
+
reasonCode: "fast-mode-disabled" | "provider-not-supported" | "fast-request-pending" | "fast-request-observed";
|
|
23
|
+
};
|
|
24
|
+
|
|
25
|
+
type SessionState = {
|
|
26
|
+
enabled: boolean;
|
|
27
|
+
source: FastModeSource;
|
|
28
|
+
observedRequestServiceTier: "priority" | null;
|
|
29
|
+
provider: string | null;
|
|
30
|
+
modelId: string | null;
|
|
31
|
+
observationPersisted: boolean;
|
|
32
|
+
};
|
|
33
|
+
|
|
34
|
+
type StateEntry = { type?: unknown; customType?: unknown; data?: unknown };
|
|
35
|
+
|
|
36
|
+
const SERVICE_TIER_RECEIPT_KEYS = [
|
|
37
|
+
"schemaVersion", "enabled", "mode", "source", "requestedServiceTier", "observedRequestServiceTier",
|
|
38
|
+
"providerResponseServiceTier", "providerResponseEvidence", "provider", "modelId", "applied", "reasonCode"
|
|
39
|
+
] as const;
|
|
40
|
+
const FAST_MODE_SOURCES = new Set<FastModeSource>(["default", "environment", "session-command", "session-state"]);
|
|
41
|
+
|
|
42
|
+
function record(value: unknown): Record<string, unknown> | null {
|
|
43
|
+
return value !== null && typeof value === "object" && !Array.isArray(value) ? value as Record<string, unknown> : null;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
function sessionId(ctx: ExtensionContext): string {
|
|
47
|
+
return ctx.sessionManager.getSessionId();
|
|
48
|
+
}
|
|
49
|
+
|
|
50
|
+
function environmentOverride(value: unknown): boolean | null {
|
|
51
|
+
if (typeof value !== "string" || !value.trim()) return null;
|
|
52
|
+
const normalized = value.trim().toLowerCase();
|
|
53
|
+
if (["1", "true", "yes", "on", "fast"].includes(normalized)) return true;
|
|
54
|
+
if (["0", "false", "no", "off", "standard", "default"].includes(normalized)) return false;
|
|
55
|
+
return null;
|
|
56
|
+
}
|
|
57
|
+
|
|
58
|
+
function restoredState(ctx: ExtensionContext): boolean | null {
|
|
59
|
+
let entries: StateEntry[] = [];
|
|
60
|
+
try { entries = ctx.sessionManager.getBranch() as StateEntry[]; } catch { return null; }
|
|
61
|
+
for (let index = entries.length - 1; index >= 0; index -= 1) {
|
|
62
|
+
const entry = entries[index], data = record(entry.data);
|
|
63
|
+
if (entry.type !== "custom" || entry.customType !== PIAGENT_FAST_MODE_STATE_ENTRY_TYPE || data?.schemaVersion !== 1) continue;
|
|
64
|
+
if (typeof data.enabled === "boolean") return data.enabled;
|
|
65
|
+
}
|
|
66
|
+
return null;
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
function isOpenAiCodex(provider: unknown): boolean {
|
|
70
|
+
return String(provider ?? "") === "openai-codex";
|
|
71
|
+
}
|
|
72
|
+
|
|
73
|
+
function boundedIdentity(value: unknown): value is string | null {
|
|
74
|
+
return value === null || typeof value === "string" && value.length <= 512 && !value.includes("\0");
|
|
75
|
+
}
|
|
76
|
+
|
|
77
|
+
export function parseServiceTierReceipt(value: unknown): ServiceTierReceipt | null {
|
|
78
|
+
const candidate = record(value);
|
|
79
|
+
if (!candidate || Object.keys(candidate).length !== SERVICE_TIER_RECEIPT_KEYS.length
|
|
80
|
+
|| !SERVICE_TIER_RECEIPT_KEYS.every((key) => Object.hasOwn(candidate, key))) return null;
|
|
81
|
+
if (candidate.schemaVersion !== 1 || typeof candidate.enabled !== "boolean"
|
|
82
|
+
|| !["standard", "fast"].includes(String(candidate.mode))
|
|
83
|
+
|| !FAST_MODE_SOURCES.has(candidate.source as FastModeSource)
|
|
84
|
+
|| ![null, PIAGENT_FAST_SERVICE_TIER].includes(candidate.requestedServiceTier as null | "fast")
|
|
85
|
+
|| ![null, PIAGENT_FAST_PROVIDER_SERVICE_TIER].includes(candidate.observedRequestServiceTier as null | "priority")
|
|
86
|
+
|| candidate.providerResponseServiceTier !== null || candidate.providerResponseEvidence !== "unavailable-host-api"
|
|
87
|
+
|| !boundedIdentity(candidate.provider) || !boundedIdentity(candidate.modelId)
|
|
88
|
+
|| typeof candidate.applied !== "boolean"
|
|
89
|
+
|| !["fast-mode-disabled", "provider-not-supported", "fast-request-pending", "fast-request-observed"].includes(String(candidate.reasonCode))) return null;
|
|
90
|
+
|
|
91
|
+
const receipt = candidate as ServiceTierReceipt;
|
|
92
|
+
const expectedMode = receipt.enabled ? "fast" : "standard";
|
|
93
|
+
const expectedRequested = receipt.enabled ? PIAGENT_FAST_SERVICE_TIER : null;
|
|
94
|
+
const supported = isOpenAiCodex(receipt.provider);
|
|
95
|
+
const expectedReason = !receipt.enabled ? "fast-mode-disabled" : !supported ? "provider-not-supported"
|
|
96
|
+
: receipt.observedRequestServiceTier === PIAGENT_FAST_PROVIDER_SERVICE_TIER ? "fast-request-observed" : "fast-request-pending";
|
|
97
|
+
const expectedApplied = receipt.enabled && supported
|
|
98
|
+
&& receipt.observedRequestServiceTier === PIAGENT_FAST_PROVIDER_SERVICE_TIER;
|
|
99
|
+
if (receipt.mode !== expectedMode || receipt.requestedServiceTier !== expectedRequested
|
|
100
|
+
|| receipt.reasonCode !== expectedReason || receipt.applied !== expectedApplied
|
|
101
|
+
|| (!receipt.enabled || !supported) && receipt.observedRequestServiceTier !== null) return null;
|
|
102
|
+
return { ...receipt };
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
export class ServiceTierRuntime {
|
|
106
|
+
readonly #environmentValue: string | undefined;
|
|
107
|
+
readonly #states = new Map<string, SessionState>();
|
|
108
|
+
|
|
109
|
+
constructor(options: { environmentValue?: string } = {}) {
|
|
110
|
+
this.#environmentValue = options.environmentValue;
|
|
111
|
+
}
|
|
112
|
+
|
|
113
|
+
restore(ctx: ExtensionContext): ServiceTierReceipt {
|
|
114
|
+
const environment = environmentOverride(this.#environmentValue);
|
|
115
|
+
const restored = environment === null ? restoredState(ctx) : null;
|
|
116
|
+
const enabled = environment ?? restored ?? false;
|
|
117
|
+
const source: FastModeSource = environment !== null ? "environment" : restored !== null ? "session-state" : "default";
|
|
118
|
+
this.#states.set(sessionId(ctx), {
|
|
119
|
+
enabled, source, observedRequestServiceTier: null, provider: null, modelId: null, observationPersisted: false
|
|
120
|
+
});
|
|
121
|
+
return this.receipt(ctx);
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
forget(ctx: ExtensionContext): void {
|
|
125
|
+
this.#states.delete(sessionId(ctx));
|
|
126
|
+
}
|
|
127
|
+
|
|
128
|
+
set(pi: ExtensionAPI, ctx: ExtensionContext, enabled: boolean): ServiceTierReceipt {
|
|
129
|
+
const state: SessionState = {
|
|
130
|
+
enabled, source: "session-command", observedRequestServiceTier: null, provider: null, modelId: null, observationPersisted: false
|
|
131
|
+
};
|
|
132
|
+
this.#states.set(sessionId(ctx), state);
|
|
133
|
+
pi.appendEntry(PIAGENT_FAST_MODE_STATE_ENTRY_TYPE, {
|
|
134
|
+
schemaVersion: 1,
|
|
135
|
+
enabled,
|
|
136
|
+
requestedServiceTier: enabled ? PIAGENT_FAST_SERVICE_TIER : null,
|
|
137
|
+
source: state.source
|
|
138
|
+
});
|
|
139
|
+
return this.receipt(ctx);
|
|
140
|
+
}
|
|
141
|
+
|
|
142
|
+
receipt(ctx: ExtensionContext): ServiceTierReceipt {
|
|
143
|
+
const state = this.#state(ctx);
|
|
144
|
+
const provider = typeof ctx.model?.provider === "string" ? ctx.model.provider : null;
|
|
145
|
+
const modelId = typeof ctx.model?.id === "string" ? ctx.model.id : null;
|
|
146
|
+
const supported = isOpenAiCodex(provider);
|
|
147
|
+
const observationMatchesCurrentModel = state.observedRequestServiceTier === PIAGENT_FAST_PROVIDER_SERVICE_TIER
|
|
148
|
+
&& state.provider === provider && state.modelId === modelId;
|
|
149
|
+
const observed = observationMatchesCurrentModel ? PIAGENT_FAST_PROVIDER_SERVICE_TIER : null;
|
|
150
|
+
return {
|
|
151
|
+
schemaVersion: 1,
|
|
152
|
+
enabled: state.enabled,
|
|
153
|
+
mode: state.enabled ? "fast" : "standard",
|
|
154
|
+
source: state.source,
|
|
155
|
+
requestedServiceTier: state.enabled ? PIAGENT_FAST_SERVICE_TIER : null,
|
|
156
|
+
observedRequestServiceTier: observed,
|
|
157
|
+
providerResponseServiceTier: null,
|
|
158
|
+
providerResponseEvidence: "unavailable-host-api",
|
|
159
|
+
provider,
|
|
160
|
+
modelId,
|
|
161
|
+
applied: state.enabled && supported && observed === PIAGENT_FAST_PROVIDER_SERVICE_TIER,
|
|
162
|
+
reasonCode: !state.enabled ? "fast-mode-disabled" : !supported ? "provider-not-supported"
|
|
163
|
+
: observed === PIAGENT_FAST_PROVIDER_SERVICE_TIER ? "fast-request-observed" : "fast-request-pending"
|
|
164
|
+
};
|
|
165
|
+
}
|
|
166
|
+
|
|
167
|
+
applyProviderRequest(pi: ExtensionAPI, ctx: ExtensionContext, payloadValue: unknown): {
|
|
168
|
+
payload: unknown;
|
|
169
|
+
changed: boolean;
|
|
170
|
+
receipt: ServiceTierReceipt;
|
|
171
|
+
observationChanged: boolean;
|
|
172
|
+
} {
|
|
173
|
+
const state = this.#state(ctx), payload = record(payloadValue);
|
|
174
|
+
if (!state.enabled || !isOpenAiCodex(ctx.model?.provider) || !payload) {
|
|
175
|
+
return { payload: payloadValue, changed: false, receipt: this.receipt(ctx), observationChanged: false };
|
|
176
|
+
}
|
|
177
|
+
const next = payload.service_tier === PIAGENT_FAST_PROVIDER_SERVICE_TIER
|
|
178
|
+
? payloadValue
|
|
179
|
+
: { ...payload, service_tier: PIAGENT_FAST_PROVIDER_SERVICE_TIER };
|
|
180
|
+
const observationChanged = state.observedRequestServiceTier !== PIAGENT_FAST_PROVIDER_SERVICE_TIER
|
|
181
|
+
|| state.provider !== String(ctx.model?.provider ?? "") || state.modelId !== String(ctx.model?.id ?? "");
|
|
182
|
+
state.observedRequestServiceTier = PIAGENT_FAST_PROVIDER_SERVICE_TIER;
|
|
183
|
+
state.provider = String(ctx.model?.provider ?? "") || null;
|
|
184
|
+
state.modelId = String(ctx.model?.id ?? "") || null;
|
|
185
|
+
const receipt = this.receipt(ctx);
|
|
186
|
+
if (observationChanged && !state.observationPersisted) {
|
|
187
|
+
pi.appendEntry(PIAGENT_SERVICE_TIER_RECEIPT_ENTRY_TYPE, receipt);
|
|
188
|
+
state.observationPersisted = true;
|
|
189
|
+
}
|
|
190
|
+
return { payload: next, changed: next !== payloadValue, receipt, observationChanged };
|
|
191
|
+
}
|
|
192
|
+
|
|
193
|
+
#state(ctx: ExtensionContext): SessionState {
|
|
194
|
+
const id = sessionId(ctx), found = this.#states.get(id);
|
|
195
|
+
if (found) return found;
|
|
196
|
+
this.restore(ctx);
|
|
197
|
+
return this.#states.get(id)!;
|
|
198
|
+
}
|
|
199
|
+
}
|
|
200
|
+
|
|
201
|
+
function receiptText(receipt: ServiceTierReceipt): string {
|
|
202
|
+
return [
|
|
203
|
+
`fastMode: ${receipt.mode}`,
|
|
204
|
+
`provider/model: ${receipt.provider && receipt.modelId ? `${receipt.provider}/${receipt.modelId}` : "unknown"}`,
|
|
205
|
+
`requestedServiceTier: ${receipt.requestedServiceTier ?? "no Piagent override (host/provider configuration applies)"}`,
|
|
206
|
+
`observedRequestServiceTier: ${receipt.observedRequestServiceTier ?? "not-observed"}`,
|
|
207
|
+
`providerMapping: Fast mode uses OpenAI service_tier=priority`,
|
|
208
|
+
`providerResponseServiceTier: unavailable (Pi host API does not expose response.service_tier)`,
|
|
209
|
+
`state: ${receipt.reasonCode}`,
|
|
210
|
+
"modelThinkingParity: unchanged",
|
|
211
|
+
"note: Fast service can consume provider quota or cost differently; Fast mode never changes model or thinking"
|
|
212
|
+
].join("\n");
|
|
213
|
+
}
|
|
214
|
+
|
|
215
|
+
export function registerFastModeCommand(pi: ExtensionAPI, runtime: ServiceTierRuntime): void {
|
|
216
|
+
pi.registerCommand("fast", {
|
|
217
|
+
description: "Show or switch OpenAI Codex Fast service tier without changing model/thinking",
|
|
218
|
+
getArgumentCompletions: (prefix: string) => ["status", "on", "off"]
|
|
219
|
+
.filter((value) => value.startsWith(String(prefix ?? "").trim().toLowerCase()))
|
|
220
|
+
.map((value) => ({ value, label: value })),
|
|
221
|
+
handler: async (args, ctx) => {
|
|
222
|
+
const action = String(args ?? "").trim().toLowerCase() || "status";
|
|
223
|
+
if (!["status", "on", "off"].includes(action)) {
|
|
224
|
+
ctx.ui.notify("Usage: /fast status|on|off", "warning");
|
|
225
|
+
return;
|
|
226
|
+
}
|
|
227
|
+
let receipt = action === "status" ? runtime.receipt(ctx) : runtime.set(pi, ctx, action === "on");
|
|
228
|
+
if (action === "on" && receipt.reasonCode === "provider-not-supported") {
|
|
229
|
+
ctx.ui.notify("Fast mode is enabled for the session, but the active provider is not OpenAI Codex.", "warning");
|
|
230
|
+
} else {
|
|
231
|
+
ctx.ui.notify(`Piagent Fast override: ${receipt.enabled ? "on" : "off"}`, receipt.enabled ? "info" : "warning");
|
|
232
|
+
}
|
|
233
|
+
receipt = runtime.receipt(ctx);
|
|
234
|
+
pi.sendMessage({
|
|
235
|
+
customType: "piagent-service-tier-receipt",
|
|
236
|
+
content: receiptText(receipt),
|
|
237
|
+
display: true,
|
|
238
|
+
details: receipt
|
|
239
|
+
}, { triggerTurn: false });
|
|
240
|
+
}
|
|
241
|
+
});
|
|
242
|
+
}
|
|
@@ -5,18 +5,19 @@ import type { RuntimeModelSnapshot } from "../model/runtime-snapshot.ts";
|
|
|
5
5
|
import type { SolverDecision, TaskFeatures } from "../solver/solver-types.ts";
|
|
6
6
|
import { bindRole, type RoleBinding } from "./role-binder.ts";
|
|
7
7
|
import { AUTOMATIC_OWNED_WORK_CEILINGS, OwnedWorkBudgetController } from "./owned-work-budget.ts";
|
|
8
|
+
import { assessMinimalDelegation, type MinimalDelegationEvidence } from "./minimal-delegation-policy.ts";
|
|
8
9
|
import { createHelperRequest, defaultRolePolicy, type HelperRequest, type HelperRole } from "./role-policy.ts";
|
|
9
10
|
|
|
10
11
|
export const HELPERS_MODE_VALUES = Object.freeze(["off", "recommend", "on"] as const);
|
|
11
12
|
export type HelpersMode = typeof HELPERS_MODE_VALUES[number];
|
|
12
|
-
export type HelperDecision = { mode: HelpersMode; action: "solo" | "recommend" | "dispatch" | "unavailable" | "blocked"; role: HelperRole | null; reasonCodes: string[]; binding: RoleBinding | null; request: HelperRequest | null };
|
|
13
|
+
export type HelperDecision = { mode: HelpersMode; action: "solo" | "recommend" | "dispatch" | "unavailable" | "blocked"; role: HelperRole | null; reasonCodes: string[]; projectedSavingsRatio: number | null; binding: RoleBinding | null; request: HelperRequest | null };
|
|
13
14
|
export type HelperDispatchResult = { status: "succeeded" | "failed" | "timeout" | "cancelled"; calls: number; tokens: number; output: string; summary?: string };
|
|
14
|
-
export type HelperUsageReceipt = { role: HelperRole; disposition: string; reasonCodes: string[]; requestRef: string; outputDigest: string | null; summary: string | null; mergeOwner: "parent" | null; calls: number; tokens: number; helperUsed: boolean };
|
|
15
|
+
export type HelperUsageReceipt = { role: HelperRole; disposition: string; decision: "dispatch" | "skip"; reasonCodes: string[]; projectedSavingsRatio: number | null; requestRef: string; outputDigest: string | null; summary: string | null; mergeOwner: "parent" | null; calls: number; tokens: number; helperUsed: boolean };
|
|
15
16
|
type DispatchAdapter = (request: HelperRequest, signal: AbortSignal) => Promise<HelperDispatchResult>;
|
|
16
17
|
type DispatchOptions = { timeoutMs?: number };
|
|
17
18
|
|
|
18
19
|
function digest(value: unknown): string { return crypto.createHash("sha256").update(String(value ?? "")).digest("hex"); }
|
|
19
|
-
export function helpersMode(value = process.env.PIAGENT_HELPERS_MODE): HelpersMode { return HELPERS_MODE_VALUES.includes(value as HelpersMode) ? value as HelpersMode : "
|
|
20
|
+
export function helpersMode(value = process.env.PIAGENT_HELPERS_MODE): HelpersMode { return HELPERS_MODE_VALUES.includes(value as HelpersMode) ? value as HelpersMode : "off"; }
|
|
20
21
|
|
|
21
22
|
export function selectHelperRole(input: { features: TaskFeatures; solver: SolverDecision; confidence?: string; conflictingEvidence?: boolean; repeatedSourceFailure?: boolean; independentReviewUseful?: boolean }): { role: HelperRole | null; reasons: string[] } {
|
|
22
23
|
if (input.features.riskLane === "tiny") return { role: null, reasons: ["tiny-task-solo"] };
|
|
@@ -31,20 +32,23 @@ export class HelperLifecycleRuntime {
|
|
|
31
32
|
readonly #budgets: OwnedWorkBudgetController;
|
|
32
33
|
readonly #active = new Map<string, Set<AbortController>>();
|
|
33
34
|
constructor(budgets = new OwnedWorkBudgetController()) { this.#budgets = budgets; }
|
|
34
|
-
decide(input: { mode: HelpersMode; objective: string; taskId: string; taskRunId: string; sessionId: string; taskScope: string[]; parentAllowedTools: string[]; features: TaskFeatures; solver: SolverDecision; runtime: RuntimeModelSnapshot; catalog: AuthenticatedModelCatalog; confidence?: string; conflictingEvidence?: boolean; repeatedSourceFailure?: boolean; independentReviewUseful?: boolean }): HelperDecision {
|
|
35
|
-
if (input.mode === "off") return { mode: input.mode, action: "solo", role: null, reasonCodes: ["helpers-off"], binding: null, request: null };
|
|
36
|
-
const selection = selectHelperRole(input); if (!selection.role) return { mode: input.mode, action: "solo", role: null, reasonCodes: selection.reasons, binding: null, request: null };
|
|
35
|
+
decide(input: { mode: HelpersMode; objective: string; taskId: string; taskRunId: string; sessionId: string; taskScope: string[]; parentAllowedTools: string[]; features: TaskFeatures; solver: SolverDecision; runtime: RuntimeModelSnapshot; catalog: AuthenticatedModelCatalog; delegationEvidence?: MinimalDelegationEvidence; confidence?: string; conflictingEvidence?: boolean; repeatedSourceFailure?: boolean; independentReviewUseful?: boolean }): HelperDecision {
|
|
36
|
+
if (input.mode === "off") return { mode: input.mode, action: "solo", role: null, reasonCodes: ["helpers-off", "parent-direct-default"], projectedSavingsRatio: null, binding: null, request: null };
|
|
37
|
+
const selection = selectHelperRole(input); if (!selection.role) return { mode: input.mode, action: "solo", role: null, reasonCodes: selection.reasons, projectedSavingsRatio: null, binding: null, request: null };
|
|
38
|
+
const delegation = assessMinimalDelegation(input.delegationEvidence);
|
|
39
|
+
if (!delegation.eligible) return { mode: input.mode, action: "solo", role: null, reasonCodes: [...selection.reasons, ...delegation.reasonCodes], projectedSavingsRatio: delegation.projectedSavingsRatio, binding: null, request: null };
|
|
37
40
|
const policy = defaultRolePolicy(selection.role, input.taskScope);
|
|
38
41
|
const binding = bindRole({ policy, features: input.features, solver: input.solver, runtime: input.runtime, catalog: input.catalog, helperBudgetAvailable: true });
|
|
39
|
-
if (binding.disposition !== "recommended") return { mode: input.mode, action: "unavailable", role: selection.role, reasonCodes: [...selection.reasons, ...binding.reasonCodes], binding, request: null };
|
|
42
|
+
if (binding.disposition !== "recommended") return { mode: input.mode, action: "unavailable", role: selection.role, reasonCodes: [...selection.reasons, ...delegation.reasonCodes, ...binding.reasonCodes], projectedSavingsRatio: delegation.projectedSavingsRatio, binding, request: null };
|
|
40
43
|
const request = createHelperRequest({ policy, objective: input.objective, taskId: input.taskId, taskRunId: input.taskRunId, sessionId: input.sessionId, parentReadScope: input.taskScope, parentWriteScope: input.taskScope, parentAllowedTools: input.parentAllowedTools, requestedReadScope: input.taskScope, requestedWriteScope: [], model: { provider: binding.provider as string, modelId: binding.modelId as string, effort: binding.effort as string, source: binding.mappingVersion }, singleWriterOwnership: null });
|
|
41
44
|
const action = input.mode === "recommend" ? "recommend" : policy.authority === "read-only" ? "dispatch" : "blocked";
|
|
42
|
-
|
|
45
|
+
const reasonCodes = [...selection.reasons, ...delegation.reasonCodes];
|
|
46
|
+
return { mode: input.mode, action, role: selection.role, reasonCodes: action === "blocked" ? [...reasonCodes, "automatic-worker-disabled"] : reasonCodes, projectedSavingsRatio: delegation.projectedSavingsRatio, binding, request };
|
|
43
47
|
}
|
|
44
48
|
async dispatch(cwd: string, decision: HelperDecision, adapter: DispatchAdapter, options: DispatchOptions = {}): Promise<HelperUsageReceipt> {
|
|
45
|
-
if (decision.action !== "dispatch" || !decision.request || !decision.role) return { role: decision.role ?? "scout", disposition: decision.action, reasonCodes: decision.reasonCodes, requestRef: "none", outputDigest: null, summary: null, mergeOwner: null, calls: 0, tokens: 0, helperUsed: false };
|
|
49
|
+
if (decision.action !== "dispatch" || !decision.request || !decision.role) return { role: decision.role ?? "scout", disposition: decision.action, decision: "skip", reasonCodes: decision.reasonCodes, projectedSavingsRatio: decision.projectedSavingsRatio, requestRef: "none", outputDigest: null, summary: null, mergeOwner: null, calls: 0, tokens: 0, helperUsed: false };
|
|
46
50
|
const reservation = this.#budgets.reserve(cwd, decision.request, undefined, AUTOMATIC_OWNED_WORK_CEILINGS);
|
|
47
|
-
if (reservation.decision !== "reserved" || !reservation.reservationId) return { role: decision.role, disposition: reservation.reason, reasonCodes: [...decision.reasonCodes, reservation.reason], requestRef: decision.request.deduplicationKey, outputDigest: null, summary: null, mergeOwner: null, calls: 0, tokens: 0, helperUsed: false };
|
|
51
|
+
if (reservation.decision !== "reserved" || !reservation.reservationId) return { role: decision.role, disposition: reservation.reason, decision: "skip", reasonCodes: [...decision.reasonCodes, reservation.reason], projectedSavingsRatio: decision.projectedSavingsRatio, requestRef: decision.request.deduplicationKey, outputDigest: null, summary: null, mergeOwner: null, calls: 0, tokens: 0, helperUsed: false };
|
|
48
52
|
const controller = new AbortController(), active = this.#active.get(decision.request.taskRunId) ?? new Set<AbortController>();
|
|
49
53
|
active.add(controller); this.#active.set(decision.request.taskRunId, active);
|
|
50
54
|
const ceilingMs = decision.request.ceilings.timeSeconds * 1000;
|
|
@@ -66,10 +70,10 @@ export class HelperLifecycleRuntime {
|
|
|
66
70
|
const released = this.#budgets.release(cwd, decision.request, reservation.reservationId, terminal, { calls: result.calls, tokens: result.tokens, output: result.output });
|
|
67
71
|
const calls = Math.min(Math.max(0, Number(result.calls) || 0), decision.request.ceilings.calls + 1);
|
|
68
72
|
const tokens = Math.min(Math.max(0, Number(result.tokens) || 0), decision.request.contextBudget + 1);
|
|
69
|
-
if (!released.accepted) return { role: decision.role, disposition: released.status === "cancelled" ? "cancelled" : "stale-result", reasonCodes: [...decision.reasonCodes, released.reason], requestRef: decision.request.deduplicationKey, outputDigest: null, summary: null, mergeOwner: null, calls, tokens, helperUsed: true };
|
|
73
|
+
if (!released.accepted) return { role: decision.role, disposition: released.status === "cancelled" ? "cancelled" : "stale-result", decision: "dispatch", reasonCodes: [...decision.reasonCodes, released.reason], projectedSavingsRatio: decision.projectedSavingsRatio, requestRef: decision.request.deduplicationKey, outputDigest: null, summary: null, mergeOwner: null, calls, tokens, helperUsed: true };
|
|
70
74
|
const withinBudget = released.reason === "released", succeeded = result.status === "succeeded" && withinBudget;
|
|
71
75
|
const summary = succeeded && typeof result.summary === "string" ? redactSensitiveText(result.summary).text.replace(/\s+/g, " ").trim().slice(0, 1000) || null : null;
|
|
72
|
-
return { role: decision.role, disposition: withinBudget ? result.status : "budget-exceeded", reasonCodes: withinBudget ? decision.reasonCodes : [...decision.reasonCodes, released.reason], requestRef: decision.request.deduplicationKey, outputDigest: succeeded && result.output ? digest(result.output) : null, summary, mergeOwner: summary ? "parent" : null, calls, tokens, helperUsed: true };
|
|
76
|
+
return { role: decision.role, disposition: withinBudget ? result.status : "budget-exceeded", decision: "dispatch", reasonCodes: withinBudget ? decision.reasonCodes : [...decision.reasonCodes, released.reason], projectedSavingsRatio: decision.projectedSavingsRatio, requestRef: decision.request.deduplicationKey, outputDigest: succeeded && result.output ? digest(result.output) : null, summary, mergeOwner: summary ? "parent" : null, calls, tokens, helperUsed: true };
|
|
73
77
|
}
|
|
74
78
|
cancelTask(cwd: string, decision: HelperDecision, now = new Date().toISOString()): number {
|
|
75
79
|
if (!decision.request) return 0;
|
|
@@ -0,0 +1,71 @@
|
|
|
1
|
+
export const MINIMAL_DELEGATION_POLICY_VERSION = "minimal-delegation-v1" as const;
|
|
2
|
+
|
|
3
|
+
export const MINIMAL_DELEGATION_LIMITS = Object.freeze({
|
|
4
|
+
maxConcurrentHelpers: 1,
|
|
5
|
+
maxTotalHelpers: 1,
|
|
6
|
+
maxRetries: 0,
|
|
7
|
+
maxHelperCalls: 8,
|
|
8
|
+
maxTransferTokens: 2_048,
|
|
9
|
+
minimumProjectedSavingsRatio: 0.3
|
|
10
|
+
});
|
|
11
|
+
|
|
12
|
+
export type MinimalDelegationEvidence = {
|
|
13
|
+
source: "runtime-estimate";
|
|
14
|
+
independentWorkstreams: string[];
|
|
15
|
+
canRunWithoutParentOutput: boolean;
|
|
16
|
+
estimatedSoloTokens: number;
|
|
17
|
+
estimatedTotalTokensWithHelper: number;
|
|
18
|
+
transferTokens: number;
|
|
19
|
+
inheritedParentTokens: number;
|
|
20
|
+
retryOfDeterministicFailure: boolean;
|
|
21
|
+
};
|
|
22
|
+
|
|
23
|
+
export type MinimalDelegationAssessment = {
|
|
24
|
+
eligible: boolean;
|
|
25
|
+
projectedSavingsRatio: number | null;
|
|
26
|
+
reasonCodes: string[];
|
|
27
|
+
};
|
|
28
|
+
|
|
29
|
+
function boundedInteger(value: unknown, minimum: number, maximum: number): value is number {
|
|
30
|
+
return Number.isInteger(value) && Number(value) >= minimum && Number(value) <= maximum;
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
function validWorkstreams(value: unknown): value is string[] {
|
|
34
|
+
return Array.isArray(value)
|
|
35
|
+
&& value.length >= 2
|
|
36
|
+
&& value.length <= 8
|
|
37
|
+
&& new Set(value).size === value.length
|
|
38
|
+
&& value.every((item) => typeof item === "string" && /^[a-z0-9][a-z0-9:._/-]{0,127}$/i.test(item));
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
/**
|
|
42
|
+
* Helpers are an optimization, not a default reasoning layer. A helper is only
|
|
43
|
+
* eligible when the runtime can show two genuinely independent lanes, a fresh
|
|
44
|
+
* minimal handoff, and at least 30% projected token savings after merge cost.
|
|
45
|
+
* The estimate is advisory, while the one-helper ledger below is enforcing.
|
|
46
|
+
*/
|
|
47
|
+
export function assessMinimalDelegation(evidence?: MinimalDelegationEvidence): MinimalDelegationAssessment {
|
|
48
|
+
if (!evidence) return { eligible: false, projectedSavingsRatio: null, reasonCodes: ["delegation-evidence-missing", "parent-direct-default"] };
|
|
49
|
+
|
|
50
|
+
const reasons: string[] = [];
|
|
51
|
+
if (evidence.source !== "runtime-estimate") reasons.push("delegation-estimate-not-runtime-owned");
|
|
52
|
+
if (!validWorkstreams(evidence.independentWorkstreams)) reasons.push("independent-workstreams-not-proven");
|
|
53
|
+
if (evidence.canRunWithoutParentOutput !== true) reasons.push("helper-depends-on-parent-output");
|
|
54
|
+
if (!boundedInteger(evidence.estimatedSoloTokens, 1, 100_000_000)
|
|
55
|
+
|| !boundedInteger(evidence.estimatedTotalTokensWithHelper, 1, 100_000_000)) reasons.push("delegation-token-estimate-invalid");
|
|
56
|
+
if (!boundedInteger(evidence.transferTokens, 0, MINIMAL_DELEGATION_LIMITS.maxTransferTokens)) reasons.push("helper-context-transfer-too-large");
|
|
57
|
+
if (evidence.inheritedParentTokens !== 0) reasons.push("parent-history-inheritance-forbidden");
|
|
58
|
+
if (evidence.retryOfDeterministicFailure !== false) reasons.push("deterministic-helper-retry-forbidden");
|
|
59
|
+
|
|
60
|
+
const projectedSavingsRatio = boundedInteger(evidence.estimatedSoloTokens, 1, 100_000_000)
|
|
61
|
+
&& boundedInteger(evidence.estimatedTotalTokensWithHelper, 1, 100_000_000)
|
|
62
|
+
? (evidence.estimatedSoloTokens - evidence.estimatedTotalTokensWithHelper) / evidence.estimatedSoloTokens
|
|
63
|
+
: null;
|
|
64
|
+
if (projectedSavingsRatio === null || projectedSavingsRatio < MINIMAL_DELEGATION_LIMITS.minimumProjectedSavingsRatio) {
|
|
65
|
+
reasons.push("projected-token-savings-below-30pct");
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
return reasons.length
|
|
69
|
+
? { eligible: false, projectedSavingsRatio, reasonCodes: [...new Set(reasons), "parent-direct-default"] }
|
|
70
|
+
: { eligible: true, projectedSavingsRatio, reasonCodes: ["independent-workstreams-proven", "projected-token-savings-at-least-30pct", "isolated-minimal-context"] };
|
|
71
|
+
}
|
|
@@ -7,12 +7,12 @@ import type { HelperRequest, HelperRole } from "./role-policy.ts";
|
|
|
7
7
|
import { validateHelperRequest } from "./role-policy.ts";
|
|
8
8
|
|
|
9
9
|
export const OWNED_WORK_BUDGET_VERSION = "owned-work-budget-v1" as const;
|
|
10
|
-
//
|
|
11
|
-
//
|
|
12
|
-
// because the repair lane is bounded elsewhere; nothing in this file enforces it.
|
|
10
|
+
// Keep the complete ceiling shape for persisted/operator projections, but clamp
|
|
11
|
+
// every caller to one read-only helper and zero retries/writers below.
|
|
13
12
|
export type OwnedWorkCeilings = Readonly<{ maxConcurrentHelpers: number; maxTotalHelpers: number; maxScoutPasses: number; maxPlannerPasses: number; maxReviewPasses: number; maxOracleCalls: number; maxRepairPasses: number; maxWriters: number }>;
|
|
14
|
-
export const
|
|
15
|
-
export const
|
|
13
|
+
export const ABSOLUTE_OWNED_WORK_CEILINGS: OwnedWorkCeilings = Object.freeze({ maxConcurrentHelpers: 1, maxTotalHelpers: 1, maxScoutPasses: 1, maxPlannerPasses: 1, maxReviewPasses: 1, maxOracleCalls: 1, maxRepairPasses: 0, maxWriters: 0 });
|
|
14
|
+
export const DEFAULT_OWNED_WORK_CEILINGS: OwnedWorkCeilings = ABSOLUTE_OWNED_WORK_CEILINGS;
|
|
15
|
+
export const AUTOMATIC_OWNED_WORK_CEILINGS: OwnedWorkCeilings = ABSOLUTE_OWNED_WORK_CEILINGS;
|
|
16
16
|
const LOCK_WAIT_MS = 5;
|
|
17
17
|
const LOCK_WAIT_CEILING_MS = 250;
|
|
18
18
|
const STALE_LOCK_MS = 30_000;
|
|
@@ -29,16 +29,6 @@ function statePath(cwd: string, taskRunId: string): string { return path.join(cw
|
|
|
29
29
|
function lockPath(cwd: string, taskRunId: string): string { return `${statePath(cwd, taskRunId)}.lock`; }
|
|
30
30
|
function empty(request: HelperRequest, now: string): BudgetState { return { version: OWNED_WORK_BUDGET_VERSION, taskId: request.taskId, taskRunId: request.taskRunId, terminal: false, reservations: [], updatedAt: now }; }
|
|
31
31
|
function active(state: BudgetState): OwnedWorkReservation[] { return state.reservations.filter((item) => item.status === "active"); }
|
|
32
|
-
// The per-role ceiling, read from the ceilings the caller passed rather than
|
|
33
|
-
// from constants written here. These limits were real but fixed at one apiece,
|
|
34
|
-
// so `OwnedWorkCeilings` advertised six numbers that nothing ever read: a
|
|
35
|
-
// profile could raise `maxScoutPasses` and be silently ignored, while the two
|
|
36
|
-
// neighbouring fields in the same object did take effect. Half-honoured
|
|
37
|
-
// configuration is worse than none, because the half that works teaches the
|
|
38
|
-
// reader to trust the half that does not.
|
|
39
|
-
//
|
|
40
|
-
// Every default is one, so this changes no behaviour under the shipped
|
|
41
|
-
// ceilings; it makes the object mean what it says.
|
|
42
32
|
function roleLimit(role: HelperRole, ceilings: OwnedWorkCeilings): number {
|
|
43
33
|
switch (role) {
|
|
44
34
|
case "scout": case "retriever": case "researcher": return ceilings.maxScoutPasses;
|
|
@@ -51,6 +41,19 @@ function roleLimit(role: HelperRole, ceilings: OwnedWorkCeilings): number {
|
|
|
51
41
|
}
|
|
52
42
|
}
|
|
53
43
|
|
|
44
|
+
function effectiveCeilings(requested: OwnedWorkCeilings): OwnedWorkCeilings {
|
|
45
|
+
return {
|
|
46
|
+
maxConcurrentHelpers: Math.min(requested.maxConcurrentHelpers, ABSOLUTE_OWNED_WORK_CEILINGS.maxConcurrentHelpers),
|
|
47
|
+
maxTotalHelpers: Math.min(requested.maxTotalHelpers, ABSOLUTE_OWNED_WORK_CEILINGS.maxTotalHelpers),
|
|
48
|
+
maxScoutPasses: Math.min(requested.maxScoutPasses, ABSOLUTE_OWNED_WORK_CEILINGS.maxScoutPasses),
|
|
49
|
+
maxPlannerPasses: Math.min(requested.maxPlannerPasses, ABSOLUTE_OWNED_WORK_CEILINGS.maxPlannerPasses),
|
|
50
|
+
maxReviewPasses: Math.min(requested.maxReviewPasses, ABSOLUTE_OWNED_WORK_CEILINGS.maxReviewPasses),
|
|
51
|
+
maxOracleCalls: Math.min(requested.maxOracleCalls, ABSOLUTE_OWNED_WORK_CEILINGS.maxOracleCalls),
|
|
52
|
+
maxRepairPasses: 0,
|
|
53
|
+
maxWriters: 0
|
|
54
|
+
};
|
|
55
|
+
}
|
|
56
|
+
|
|
54
57
|
function read(cwd: string, request: HelperRequest, now: string): BudgetState {
|
|
55
58
|
const target = statePath(cwd, request.taskRunId); if (!fs.existsSync(target)) return empty(request, now);
|
|
56
59
|
const resolved = resolveLocalStatePath(cwd, target, { label: "Helper budget state", kind: "file" });
|
|
@@ -153,14 +156,19 @@ export function inspectOwnedWorkBudget(cwd: string, taskId: string, taskRunId: s
|
|
|
153
156
|
export class OwnedWorkBudgetController {
|
|
154
157
|
reserve(cwd: string, requestInput: HelperRequest, now = new Date().toISOString(), ceilings: OwnedWorkCeilings = DEFAULT_OWNED_WORK_CEILINGS): { decision: "reserved" | "duplicate" | "blocked"; reason: string; reservationId: string | null; recoveredOrphans: number } {
|
|
155
158
|
const request = validateHelperRequest(structuredClone(requestInput));
|
|
159
|
+
const limits = effectiveCeilings(ceilings);
|
|
156
160
|
return withLock(cwd, request.taskRunId, () => {
|
|
157
161
|
const state = read(cwd, request, now); const recoveredOrphans = recoverOrphans(state, now);
|
|
158
|
-
|
|
162
|
+
const settle = (result: { decision: "reserved" | "duplicate" | "blocked"; reason: string; reservationId: string | null; recoveredOrphans: number }) => {
|
|
163
|
+
if (recoveredOrphans > 0) { state.updatedAt = now; write(cwd, state); }
|
|
164
|
+
return result;
|
|
165
|
+
};
|
|
166
|
+
if (state.terminal) return settle({ decision: "blocked", reason: "parent-task-terminal", reservationId: null, recoveredOrphans });
|
|
159
167
|
const duplicate = state.reservations.find((item) => item.deduplicationKey === request.deduplicationKey && ["active", "succeeded"].includes(item.status));
|
|
160
|
-
if (duplicate) return { decision: "duplicate", reason: "equivalent-helper-already-owned", reservationId: duplicate.id, recoveredOrphans };
|
|
161
|
-
if (active(state).length >=
|
|
162
|
-
if (state.reservations.filter((item) => item.role === request.role).length >= roleLimit(request.role,
|
|
163
|
-
if (request.authority === "single-writer" && active(state).some((item) => item.authority === "single-writer")) return { decision: "blocked", reason: "single-writer-already-owned", reservationId: null, recoveredOrphans };
|
|
168
|
+
if (duplicate) return settle({ decision: "duplicate", reason: "equivalent-helper-already-owned", reservationId: duplicate.id, recoveredOrphans });
|
|
169
|
+
if (active(state).length >= limits.maxConcurrentHelpers || state.reservations.length >= limits.maxTotalHelpers) return settle({ decision: "blocked", reason: "helper-budget-exhausted", reservationId: null, recoveredOrphans });
|
|
170
|
+
if (state.reservations.filter((item) => item.role === request.role).length >= roleLimit(request.role, limits)) return settle({ decision: "blocked", reason: `${request.role}-ceiling-reached`, reservationId: null, recoveredOrphans });
|
|
171
|
+
if (request.authority === "single-writer" && active(state).some((item) => item.authority === "single-writer")) return settle({ decision: "blocked", reason: "single-writer-already-owned", reservationId: null, recoveredOrphans });
|
|
164
172
|
const id = digest(`${request.taskRunId}:${request.deduplicationKey}:${state.reservations.length}`).slice(0, 32);
|
|
165
173
|
state.reservations.push({ id, deduplicationKey: request.deduplicationKey, role: request.role, authority: request.authority, status: "active", reservedAt: now, expiresAt: new Date(Date.parse(now) + request.ceilings.timeSeconds * 1000).toISOString(), completedAt: null, usageRef: null });
|
|
166
174
|
state.updatedAt = now; write(cwd, state); return { decision: "reserved", reason: "budget-reserved", reservationId: id, recoveredOrphans };
|
|
@@ -3,7 +3,7 @@ import { globMatchesPath, normalizePathCandidate } from "../../extensions/policy
|
|
|
3
3
|
|
|
4
4
|
export const ROLE_POLICY_SCHEMA_VERSION = 1 as const;
|
|
5
5
|
export const ROLE_POLICY_VERSION = "role-policy-v1" as const;
|
|
6
|
-
export const HELPER_REQUEST_SCHEMA_VERSION =
|
|
6
|
+
export const HELPER_REQUEST_SCHEMA_VERSION = 2 as const;
|
|
7
7
|
export const HELPER_ROLES = Object.freeze(["retriever", "scout", "planner", "worker", "reviewer", "oracle", "researcher"] as const);
|
|
8
8
|
export type HelperRole = typeof HELPER_ROLES[number];
|
|
9
9
|
export type HelperAuthority = "read-only" | "single-writer";
|
|
@@ -15,12 +15,13 @@ export type RolePolicy = {
|
|
|
15
15
|
outputSchema: string; stoppingRule: string; approvalRestrictions: string[];
|
|
16
16
|
};
|
|
17
17
|
export type HelperRequest = {
|
|
18
|
-
schemaVersion:
|
|
18
|
+
schemaVersion: 2; policyVersion: typeof ROLE_POLICY_VERSION; role: HelperRole; objectiveHash: string; objectiveText: string;
|
|
19
19
|
taskId: string; taskRunId: string; sessionHash: string; authority: HelperAuthority; readScope: string[]; writeScope: string[];
|
|
20
20
|
allowedTools: string[]; model: { provider: string; modelId: string; effort: string; source: string } | null;
|
|
21
21
|
contextBudget: number; ceilings: { timeSeconds: number; calls: number; retries: number }; outputSchema: string;
|
|
22
22
|
stoppingRule: string; approvalRestrictions: string[]; parentAuthorityDigest: string; singleWriterOwnership: string | null;
|
|
23
23
|
deduplicationKey: string;
|
|
24
|
+
contextTransfer: { mode: "isolated-minimal"; inheritParentHistory: false; maxSeedTokens: number; estimatedSeedTokens: number };
|
|
24
25
|
};
|
|
25
26
|
|
|
26
27
|
const READ_TOOLS = ["read", "grep", "find", "ls"];
|
|
@@ -28,7 +29,7 @@ const MUTATION_TOOLS = new Set(["edit", "write", "apply_patch", "bash", "contact
|
|
|
28
29
|
const HASH = /^[a-f0-9]{64}$/;
|
|
29
30
|
const REF = /^[a-z0-9][a-z0-9:._/-]{0,255}$/i;
|
|
30
31
|
const ROLE_FIELDS = new Set(["schemaVersion", "policyVersion", "role", "authority", "enabledByDefault", "readScope", "writeScope", "allowedTools", "modelSelectionSource", "effortSelectionSource", "contextBudget", "ceilings", "outputSchema", "stoppingRule", "approvalRestrictions"]);
|
|
31
|
-
const REQUEST_FIELDS = new Set(["schemaVersion", "policyVersion", "role", "objectiveHash", "objectiveText", "taskId", "taskRunId", "sessionHash", "authority", "readScope", "writeScope", "allowedTools", "model", "contextBudget", "ceilings", "outputSchema", "stoppingRule", "approvalRestrictions", "parentAuthorityDigest", "singleWriterOwnership", "deduplicationKey"]);
|
|
32
|
+
const REQUEST_FIELDS = new Set(["schemaVersion", "policyVersion", "role", "objectiveHash", "objectiveText", "taskId", "taskRunId", "sessionHash", "authority", "readScope", "writeScope", "allowedTools", "model", "contextBudget", "ceilings", "outputSchema", "stoppingRule", "approvalRestrictions", "parentAuthorityDigest", "singleWriterOwnership", "deduplicationKey", "contextTransfer"]);
|
|
32
33
|
|
|
33
34
|
function object(value: unknown): Record<string, any> | undefined { return value && typeof value === "object" && !Array.isArray(value) ? value as Record<string, any> : undefined; }
|
|
34
35
|
function exact(value: Record<string, any>, fields: Set<string>, label: string): string[] { return [...Object.keys(value).filter((field) => !fields.has(field)).map((field) => `${label} unknown field: ${field}`), ...[...fields].filter((field) => !(field in value)).map((field) => `${label} missing field: ${field}`)]; }
|
|
@@ -36,6 +37,15 @@ function strings(value: unknown, max: number, itemMax = 300): value is string[]
|
|
|
36
37
|
function ceilings(value: unknown): boolean { const item = object(value); return Boolean(item && exact(item, new Set(["timeSeconds", "calls", "retries"]), "ceilings").length === 0 && Number.isInteger(item.timeSeconds) && item.timeSeconds >= 1 && item.timeSeconds <= 3600 && Number.isInteger(item.calls) && item.calls >= 1 && item.calls <= 100 && Number.isInteger(item.retries) && item.retries >= 0 && item.retries <= 3); }
|
|
37
38
|
function digest(value: unknown): string { return crypto.createHash("sha256").update(JSON.stringify(value)).digest("hex"); }
|
|
38
39
|
function boundedObjective(value: unknown): string { return String(value ?? "").replace(/\s+/g, " ").trim().slice(0, 1000); }
|
|
40
|
+
function estimatedSeedTokens(value: Record<string, any>): number {
|
|
41
|
+
const seed = {
|
|
42
|
+
schemaVersion: value.schemaVersion, policyVersion: value.policyVersion, role: value.role, objectiveText: value.objectiveText,
|
|
43
|
+
taskId: value.taskId, taskRunId: value.taskRunId, authority: value.authority, readScope: value.readScope, writeScope: value.writeScope,
|
|
44
|
+
allowedTools: value.allowedTools, model: value.model, contextBudget: value.contextBudget, ceilings: value.ceilings,
|
|
45
|
+
outputSchema: value.outputSchema, stoppingRule: value.stoppingRule, approvalRestrictions: value.approvalRestrictions
|
|
46
|
+
};
|
|
47
|
+
return Math.ceil(JSON.stringify(seed).length / 4);
|
|
48
|
+
}
|
|
39
49
|
function safeScopePattern(value: string): boolean {
|
|
40
50
|
const normalized = normalizePathCandidate(value);
|
|
41
51
|
return Boolean(normalized && !/^\/|^[A-Za-z]:\//.test(normalized) && !normalized.split("/").some((part) => part === "." || part === "..") && !/[?\[\]{}\0]/.test(normalized));
|
|
@@ -69,7 +79,7 @@ export function validateRolePolicy(input: unknown): RolePolicy { const errors =
|
|
|
69
79
|
export function helperRequestValidationErrors(input: unknown): string[] {
|
|
70
80
|
const value = object(input); if (!value) return ["helper request must be an object"];
|
|
71
81
|
const errors = exact(value, REQUEST_FIELDS, "helper request");
|
|
72
|
-
if (value.schemaVersion !==
|
|
82
|
+
if (value.schemaVersion !== 2 || value.policyVersion !== ROLE_POLICY_VERSION || !HELPER_ROLES.includes(value.role) || !HASH.test(String(value.objectiveHash)) || !HASH.test(String(value.sessionHash)) || !HASH.test(String(value.parentAuthorityDigest)) || !HASH.test(String(value.deduplicationKey))) errors.push("helper identity/digests are invalid");
|
|
73
83
|
if (typeof value.objectiveText !== "string" || value.objectiveText.length < 3 || value.objectiveText.length > 1000 || !REF.test(String(value.taskId)) || !REF.test(String(value.taskRunId))) errors.push("helper objective/task identity is invalid");
|
|
74
84
|
if (!["read-only", "single-writer"].includes(value.authority) || !strings(value.readScope, 100) || !strings(value.writeScope, 100) || !value.readScope.every(safeScopePattern) || !value.writeScope.every(safeScopePattern) || !strings(value.allowedTools, 32) || !strings(value.approvalRestrictions, 16)) errors.push("helper authority collections are invalid");
|
|
75
85
|
if (value.authority === "read-only" && (value.writeScope.length > 0 || value.allowedTools.some((tool: string) => MUTATION_TOOLS.has(tool)))) errors.push("read-only helper cannot mutate");
|
|
@@ -79,6 +89,12 @@ export function helperRequestValidationErrors(input: unknown): string[] {
|
|
|
79
89
|
if (typeof value.outputSchema !== "string" || value.outputSchema.length > 200 || typeof value.stoppingRule !== "string" || value.stoppingRule.length > 500) errors.push("helper output/stopping contract is invalid");
|
|
80
90
|
const model = value.model === null ? null : object(value.model);
|
|
81
91
|
if (value.model !== null && (!model || exact(model, new Set(["provider", "modelId", "effort", "source"]), "model").length || ![model.provider, model.modelId, model.effort, model.source].every((item) => typeof item === "string" && item.length > 0 && item.length <= 160))) errors.push("helper model binding is invalid");
|
|
92
|
+
const contextTransfer = object(value.contextTransfer);
|
|
93
|
+
if (!contextTransfer || exact(contextTransfer, new Set(["mode", "inheritParentHistory", "maxSeedTokens", "estimatedSeedTokens"]), "context transfer").length
|
|
94
|
+
|| contextTransfer.mode !== "isolated-minimal" || contextTransfer.inheritParentHistory !== false
|
|
95
|
+
|| !Number.isInteger(contextTransfer.maxSeedTokens) || contextTransfer.maxSeedTokens < 256 || contextTransfer.maxSeedTokens > 2048
|
|
96
|
+
|| !Number.isInteger(contextTransfer.estimatedSeedTokens) || contextTransfer.estimatedSeedTokens !== estimatedSeedTokens(value)
|
|
97
|
+
|| contextTransfer.estimatedSeedTokens > contextTransfer.maxSeedTokens) errors.push("helper context transfer must be isolated and minimal");
|
|
82
98
|
return errors;
|
|
83
99
|
}
|
|
84
100
|
export function validateHelperRequest(input: unknown): HelperRequest { const errors = helperRequestValidationErrors(input); if (errors.length) throw new Error(errors.join("; ")); return input as HelperRequest; }
|
|
@@ -87,9 +103,9 @@ export function defaultRolePolicy(role: HelperRole, scope: string[] = ["**"]): R
|
|
|
87
103
|
const readOnly = role !== "worker";
|
|
88
104
|
const policy: RolePolicy = {
|
|
89
105
|
schemaVersion: 1, policyVersion: ROLE_POLICY_VERSION, role, authority: readOnly ? "read-only" : "single-writer",
|
|
90
|
-
enabledByDefault:
|
|
91
|
-
modelSelectionSource: "runtime-catalog", effortSelectionSource: "runtime-snapshot", contextBudget: role === "oracle" ?
|
|
92
|
-
ceilings: { timeSeconds:
|
|
106
|
+
enabledByDefault: false, readScope: [...scope], writeScope: [], allowedTools: readOnly ? [...READ_TOOLS] : [...READ_TOOLS, "bash", "edit", "write", "apply_patch", "contact_supervisor"],
|
|
107
|
+
modelSelectionSource: "runtime-catalog", effortSelectionSource: "runtime-snapshot", contextBudget: role === "oracle" ? 8000 : role === "reviewer" || role === "planner" ? 6000 : 4000,
|
|
108
|
+
ceilings: { timeSeconds: 300, calls: 8, retries: 0 },
|
|
93
109
|
outputSchema: `${role}-result-v1`, stoppingRule: "Stop when the bounded objective is answered or evidence is insufficient; return uncertainty.",
|
|
94
110
|
approvalRestrictions: ["no-external-write", "no-destructive-action", "no-permission-expansion"]
|
|
95
111
|
};
|
|
@@ -103,11 +119,13 @@ export function createHelperRequest(input: { policy: RolePolicy; objective: stri
|
|
|
103
119
|
if (policy.allowedTools.some((tool) => !input.parentAllowedTools.includes(tool))) throw new Error("helper request cannot broaden parent tools");
|
|
104
120
|
const authority = policy.authority;
|
|
105
121
|
const request: HelperRequest = {
|
|
106
|
-
schemaVersion:
|
|
122
|
+
schemaVersion: 2, policyVersion: ROLE_POLICY_VERSION, role: policy.role, objectiveHash: digest(objectiveText), objectiveText,
|
|
107
123
|
taskId: input.taskId, taskRunId: input.taskRunId, sessionHash: digest(input.sessionId), authority, readScope: [...readScope], writeScope: [...writeScope], allowedTools: [...policy.allowedTools],
|
|
108
124
|
model: input.model ?? null, contextBudget: policy.contextBudget, ceilings: { ...policy.ceilings }, outputSchema: policy.outputSchema, stoppingRule: policy.stoppingRule,
|
|
109
125
|
approvalRestrictions: [...policy.approvalRestrictions], parentAuthorityDigest: digest({ read: input.parentReadScope, write: input.parentWriteScope, tools: input.parentAllowedTools }),
|
|
110
|
-
singleWriterOwnership: input.singleWriterOwnership ?? null, deduplicationKey: digest({ role: policy.role, objective: objectiveText, readScope, writeScope })
|
|
126
|
+
singleWriterOwnership: input.singleWriterOwnership ?? null, deduplicationKey: digest({ role: policy.role, objective: objectiveText, readScope, writeScope }),
|
|
127
|
+
contextTransfer: { mode: "isolated-minimal", inheritParentHistory: false, maxSeedTokens: 2048, estimatedSeedTokens: 0 }
|
|
111
128
|
};
|
|
129
|
+
request.contextTransfer.estimatedSeedTokens = estimatedSeedTokens(request);
|
|
112
130
|
return validateHelperRequest(request);
|
|
113
131
|
}
|